NEURAL-NETWORK APPROACH TO FATIGUE-CRACK-GROWTH PREDICTIONS UNDER AIRCRAFT SPECTRUM LOADINGS

An artificial neural network (NN) method is developed to represent the fatigue-crack-growth and cycle relationships under spectrum loadings of the Mirage aircraft operated by the Royal Australian Air Force. Thi s method utilizes load cycle spectrum using available flight and exper imental data for crack growth vs cycles as input. The trained network is able to predict the relationship between the crack-growth and the l oading cycles. The neural network is able to predict the crack-growth cycle behavior for different variations in the original loading spectr ums. The results predicted by the NN model seem reasonable and the mod el is capable of representing crack-growth behavior for various arbitr ary aircraft spectrum loadings with certain limitations. In addition, an attempt is made to predict the material parameters for Walker's fat igue-crack-growth relationship using a different neural network. Becau se of the demonstrated performance, it is possible that the proposed N N approach can be extended with more research effort to estimate the f atigue life of arbitrary cracked structural components under complex l oadings in real time.